Search papers, labs, and topics across Lattice.
SimSkill is a self-evolving AI agent that leverages the Simulation of Urban MObility (SUMO) traffic simulator to autonomously identify capability gaps and generate environment-grounded tasks, ultimately consolidating its experiences into various memory types without altering the backbone model. The agent's performance was evaluated against two held-out benchmarks with three different backbone large language models (LLMs), achieving up to a 25 percentage point improvement in verified task completion. Notably, the study reveals that the effectiveness of SimSkill's memory mechanisms is dependent on the specific model and resource budget used, highlighting the nuanced relationship between memory and inference efficiency.
SimSkill transforms traffic simulation mastery by autonomously evolving its capabilities and consolidating knowledge without modifying its core model.
As large language models (LLMs) become increasingly capable, the long-term value of AI systems depends not only on solving individual requests, but also on transforming experience and accumulated knowledge into durable, reusable competence. We introduce SimSkill, a self-evolving agent built around the Simulation of Urban MObility (SUMO) traffic simulator. SimSkill identifies capability gaps, generates and solves environment-grounded tasks, verifies solutions through an action--critic loop, and consolidates experience into episodic, procedural, and semantic memory without updating the backbone model. Through autonomous exploration, it builds a reusable library spanning the traffic-simulation workflow. We evaluate SimSkill on two held-out benchmarks with three backbone LLMs and independent artifact-based verification. SimSkill improves verified completion by up to 25 percentage points, while ablations show complementary contributions from procedural and semantic memory. Its benefits remain backbone- and budget-dependent: memory does not improve every model or uniformly reduce inference cost. More broadly, SimSkill illustrates a design paradigm in which natural language preserves and composes computational capabilities, while executable tools and code provide precise and reproducible execution. All code and experimental data are publicly available at https://github.com/qiliuchn/SimSkill-V1.
